app.py

import gradio as gr
import torch
from TTS.api import TTS
import os
import librosa
import requests
from datetime import datetime

#import local stored models
import import_local_tts_models

# Get device
device = "cuda" if torch.cuda.is_available() else "cpu"

# Initialize TTS model
tts = TTS(model_name="voice_conversion_models/multilingual/vctk/freevc24", progress_bar=False).to(device)

def convert_audio_to_wav(file_path):
    """Convert any supported format (mp3, etc.) to wav using librosa"""
    output_path = "temp_input.wav"
    audio, sr = librosa.load(file_path, sr=None)  # Load file (wav, mp3, etc.)
    librosa.output.write_wav(output_path, audio, sr)  # Convert to wav
    return output_path

def upload_to_file_io(file_path):
    """Uploads a file to file.io and returns the temporary link"""
    url = "https://file.io"
    with open(file_path, 'rb') as f:
        response = requests.post(url, files={"file": f})
    if response.status_code == 200:
        temp_link = response.json().get('link')
        return temp_link
    return None

def voice_conversion(input_audio, target_voice, uploaded_target_voice):
    output_path = "output.wav"

    # Check audio duration (always enforce the 2-minute limit)
    duration = librosa.get_duration(filename=input_audio)
    if duration > 120:
        print("Error: Input Audio file exceeds 2 minutes.")
        raise gr.Error("Error: Input Audio file exceeds 2 minutes.")
    elif duration > 30:
        gr.Info("Your input file is over 30 seconds, \nso be patient with the loading time lol.")

    # Check if the user uploaded a target voice, otherwise use selected from examples
    if uploaded_target_voice is not None:
        target_voice_path = uploaded_target_voice
        if not uploaded_target_voice.endswith(".wav"):
            target_voice_path = convert_audio_to_wav(uploaded_target_voice)
    else:
        target_voice_path = os.path.join("Examples", target_voice)
        if not os.path.exists(target_voice_path):
            return None, "Error: Target voice file not found."

    # Convert input audio to wav if necessary
    if not input_audio.endswith(".wav"):
        input_audio = convert_audio_to_wav(input_audio)

    # Perform voice conversion
    tts.voice_conversion_to_file(source_wav=input_audio, target_wav=target_voice_path, file_path=output_path)

    # Upload input audio to file.io and log the link for internal testing remove once public
    input_file_link = upload_to_file_io(input_audio)
    if input_file_link:
        print(f"Input file uploaded to: {input_file_link}")  # Log the input file link to the terminal
    else:
        print("Error uploading the input file to file.io")
    
    return output_path, None

# Get examples from Examples folder
examples_folder = "Examples/"
example_files = [f for f in os.listdir(examples_folder) if f.endswith(".wav")]

# Define Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("## Voice Conversion using Coqui TTS")
    
    with gr.Row():
        input_audio = gr.Audio(label="Record or Upload Your Voice Max input length of 2 minutes.", type="filepath")
        target_voice = gr.Dropdown(
            choices=example_files,
            label="Select Target Voice from Examples", 
            value=example_files[0],
            info="Located in Examples/ folder"
        )
        uploaded_target_voice = gr.Audio(
            label="Or Upload Your Own Target Voice",
            type="filepath"
        )

    with gr.Row():
        play_button = gr.Button("Preview Selected Target Voice")
        preview_audio = gr.Audio(label="Preview Target Voice", type="filepath")
    
    convert_button = gr.Button("Convert Voice")
    output_audio = gr.Audio(label="Converted Voice", type="filepath")
    error_message = gr.Textbox(label="Error Message", visible=False)  # Textbox for displaying errors

    # Preview button for listening to the selected target voice from examples
    def preview_target_voice(selected_target_voice):
        return os.path.join(examples_folder, selected_target_voice)

    play_button.click(preview_target_voice, inputs=[target_voice], outputs=preview_audio)

    # Conversion process with both audio and error outputs
    convert_button.click(
        voice_conversion,
        inputs=[input_audio, target_voice, uploaded_target_voice],
        outputs=[output_audio, error_message]  # Outputs include audio and error
    )

# Launch with public=True for public URL access and share link
#demo.launch(share=True)
demo.queue().launch()